FEAT 3 - Advanced FMRI FEAT 3 - Advanced FMRI Analysis Pipeline overview Advanced preprocessing steps

  • View
    4

  • Download
    0

Embed Size (px)

Text of FEAT 3 - Advanced FMRI FEAT 3 - Advanced FMRI Analysis Pipeline overview Advanced preprocessing...

  • FEAT 3 - Advanced FMRI Analysis

    Pipeline overview Advanced preprocessing steps

    • Motion artefact correction

    • Physiological noise correction

    Demeaning EVs

    Advanced designs: • Parametric designs and F-

    tests • Factorial designs and

    interactions • Contrast masking

    • Correlated EVs • Design efficiency • F-tests

  • Pipeline overview

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Aims: ●Obtain good

    quality and consistent data

    ●Optimise SNR

    Keep in mind: ●Many trade-

    offs ●Consider

    drop-out and distortions

    ●What are the most important regions?

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Aims: ●Reduce noise

    in data ●Prepare data

    for analysis ●Prepare data

    for group comparison

    Keep in mind: ●Requires

    careful checking

    ●Can add additional steps if necessary

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Aims: ●Obtain

    measure of interest for each subject (often an image)

    Keep in mind: ●Differs

    considerably between modalities

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Aims: ●Compare

    single-subject results across group

    ●Group mean/ t-test/ correlation

    Keep in mind: ●Can have

    additional layer to average over sessions

    ●Account for confounding variables

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Aims: ●P-values ●Reliability of

    results ●Generalise to

    population

    Keep in mind: ●Need enough

    subjects to have power

    ●Cannot interpret null results

  • What we covered so far

    Brain extraction

    Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    VBM or vertex analysis

  • Preprocessing

    Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    Brain extraction

    VBM or vertex analysis

  • Brain extraction Remove non-brain tissue to help with registration. Needs to be very precise.

    Bias field correction Corrects for B1 inhomogeneities

    Registration Put images into same space (standard space for group analysis)

    Structural preprocessing summary

  • fMRI preprocessing summary

    Brain extraction Remove non-brain tissue to help with registration

    Motion Correction Get consistent anatomical coordinates (always do this)

    Slice Timing Get consistent acquisition timing (use temporal derivative instead)

    Spatial Smoothing Improve SNR & validate GRF

    Temporal Filtering Highpass: Remove slow drifts

    Registration & unwarping Unwarping corrects for B0 inhomogeneities. Registration images into same space (standard space for group analysis)

  • Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    VBM or vertex analysis

    Brain extraction

    Single-subject analysis

  • Structural single-subject summary

    Segmentation Tissue-type segmentation (FAST), sub- cortical segmentation (FIRST), white matter hyperintensities (BIANCA)

    Voxel-based morphometry To detect differences in local grey matter volume. Jacobian modulation and spatial smoothing.

    Vertex analysis To run shape analysis on subcortical structures. first_utils uses bvars output from FIRST to perform vertex analysis (4D output image of all subject meshes)

  • fMRI single-subject summary

    EVs/ regressors Design matrix: model of predicted responses based on stimuli presented at each time point

    GLM Estimate parameter estimates for each EV so that the linear combination best fits the data

    Contrasts (F or t) Maths on parameter estimates to ask research questions. Result is a COPE image per contrast

  • Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    VBM or vertex analysis

    Brain extraction

    Group-level analysis

  • Group-level analysis summary

    EVs/ regressors Design matrix: one entry per subject. Can describe subject groups, confounds etc

    GLM Structural: inputs are smoothed, modulated GM volumes (VBM) or single subject subcortical meshes (vertex analysis)

    fMRI: inputs are first-level COPE and VARCOPE images.

    Contrasts (F or t) Structural: tests differences in GM density or shape fMRI: Each group-level contrast is tested for each of the subject-level contrasts

  • Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    VBM or vertex analysis

    Brain extraction

    Statistical inference

  • Statistical inference summary

    Fixed effects vs

    mixed effects

    Averaging across multiple sessions

    Generalisation to population

    OLS vs

    FLAME vs Randomise

    Quick, doesn’t use VARCOPEs

    Uses COPEs & VARCOPEs

    Non-parametric

    Multiple comparison correction (FWE/ FDR)

    Gaussian Random Field (voxel or cluster based)

    TFCE

  • Motion correction

    Slice timing correction

    Spatial filtering

    Temporal filtering

    Registration & unwarping

    Structural data:

    Functional data:

    Regressors & contrasts

    First level GLM

    Group level GLM

    Thresholding & correction

    Regressors & contrasts

    Bias field correction Segmentation

    VBM or vertex analysis

    Brain extraction

    Preprocessing

    Single-subject analysis

    Group-level analysis Statistical inference

    What we covered so far

  • Looking ahead: resting state

    diffusion arterial spin labeling

  • Generic blueprint

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

  • Resting state analysis

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Consider using multiband

  • Resting state analysis

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Need to apply extra noise- reduction steps (ICA)

    Consider using multiband

  • Resting state analysis

    1. Data acquisition

    2. Data preprocessing

    3. Single-subject analysis

    4. Group-level analysis

    5. Statistical inference

    Group ICA+dual regression/ Network analysis (FSLnets)

    Consider using multiband

    Need to apply extra noise- reduction steps (ICA)

  • Diffusion analysis

    1. Data acquis